The storage systems of International Business Machines Inc. (IBM) and its plug compatible imitators are first described along with IBM's overall storage strategy. The place for Write Once Read Many (WORM) optical storage of the present invention is then identified and characterized with possible optical implementations including benefits and disadvantages. IBM mainframe computers have an Input/Output Channel, referred to at times as the I/O Channel or simply the Channel, to which the peripheral devices attach for communication with the computer. The attached peripheral typically include high capacity magnetic tape and disk drives.
The assignee of the present invention, (Data/Ware Development Inc. San Diego, Calif.), first offered an I/O Channel Tester as an IBM channel attached product. The I/O Channel Tester was needed to simulate peripherals in order to debug and test IBM I/O channels. The I/O Channel Tester is unique in that it forces errors on command. The I/O Channel Tester is a generic control unit emulator. The I/O Channel Tester was designed for the plug compatible mainframe manufacturers. The unit is able to emulate the operation of byte/block/selector type controllers from ten kilo bytes per second to three mega bytes per second.
The Peripheral Automatic Channel Emulator (PACE) is assignees's second channel product. PACE emulates an IBM I/O Channel using a personal computer as its host. PACE allows peripheral manufacturers to do development and test without an expensive mainframe. PACE is capable of data streaming transfers at three mega bytes per second. PACE is a mainframe-in-a-box used by plug compatible peripheral developers and manufacturers in place of mainframe I/O channels. PACE attaches to an IBM personal computer and provides full channel emulation, byte/block/selector up to three or four-point-five mega bytes per second. In addition to the PACE, diagnostic programs for standard peripherals including 3211, 3420, 3480, 3380 have been developed. A program equivalent to IBM's FRIEND diagnostic called PAL has also been developed.
The Channel Monitor is the third channel attached product developed by the Assignee. Like a logic analyzer, the Channel Monitor is able to capture and display the sequences used to control a peripheral and to transfer data. The Channel Monitor is an intelligent tool, attaches directly to the bus and tag cables, and is preprogrammed to understand the protocol of the I/O Channel. This makes it easy to use for field service and installation test. The Channel Monitor is a specialized logic analyzer for the IBM I/O channel. This device is also useful for characterization of control units.
The Assignee's fourth product is the VMEGate and evolved from the IBM I/O Channel test equipment. Combining the industry standard VME bus with the bit-slice channel technology evident in the I/O Channel Tester and PACE, the VMEGate was the first system providing a flexible, user programmable channel attachment mechanism for Original Equipment Manufactures. The family of VMEGates includes both channel and control unit emulations.
The Assignee's fifth product, the PCGate, is similar to VMEGate in that it makes channel attachment available on a standard bus. In this case, the channel attachment is to an IBM PC bus. PCGate permits attachment of an IBM PC to a mainframe computer directly at the bus and tag level. This results in a dramatic improvement in transfer speed, from the fifty-six kilo bits per second available with the previous SDLC techniques to three hundred kilo bytes per second.
Plug compatible peripheral designs for the IBM I/O Channel require IBM I/O Channel experience and knowledge and the importance of reliability and conformance to IBM conventions. Incompatibility with different mainframe models, different operating systems and various third party / peripherals have frustrated the mainframe users. The present invention is primarily focused on the IBM mainframe computer direct-channel-attached marketplace. The following section describes the IBM storage hierarchy and the individual items of equipment which make up this hierarchy. This establishes the baseline for the introduction of optical storage and the present invention.
The mainframe storage hierarchy consists i) of main memory, ii) auxiliary memory such as cache and solid state disk, iii) rotating disk, iv) reel and cartridge tape, and v) hardcopy such as microfilm and paper. This hierarchy is arranged according to functionality, i.e. the upper level storage mechanisms are the most accessible. In general, the upper levels are also the most expensive, the fastest, and the most volatile both by nature and by design. Any new mechanism must find its place in this existing hierarchy.
The top of the storage hierarchy is occupied by semiconductor memory directly addressable by the computer's CPU (central processing unit). The original IBM 370 architecture used a twenty-four bit addressing scheme providing access to sixteen megabytes of real memory. This became a serious constraint on larger systems, and led to the introduction of the IBM XA computer. The IBM Extended Architecture XA computers provide a thirty-one bit address providing access to two-point-one giga bytes. This expansion of CPU memory size was necessary to balance the increase in executed instructions per second. The advent of more powerful central processors necessitated the addition of more available memory.
The original IBM Operating System (OS) was a real storage operating system. In 1972 IBM produced the OS/SVS operating system which provided Single Virtual Storage (SVS) addressing for the first time. In 1974, IBM introduced the OS/MVS operating system having Multiple Virtual Storage (MVS) addressing on the IBM 370 computer which then provided multiple virtual storage address spaces of sixteen megabytes to each of several concurrent system users. As larger processors came online it became more difficult to balance an MVS system. The XA computer is designed to better utilize resources and make it easier for users to balance work loads. The Extended Architecture, XA, computer is a combination of System-370/XA hardware and MVS/XA software designed to efficiently manage the system resources of the processor, virtual storage, real storage, and I/O. Auxiliary memory is semiconductor memory accessible over the I/O Channel. This includes solid state disk and cache. This memory is accessed through I/O instructions and is consequently less accessible than main memory. Because there are no mechanical delays (e.g. seek delay) associated with auxiliary memory, it is faster than other I/O devices. Cache is designed to keep frequently used data close at hand. The size of cache has been increasing and a cache speed increase is easier to achieve than a speed increase directly from the disk. Cache can eliminate rotational latency and speed access to subsequent datasets much like the solid state disk. The solid state disk uses semiconductor memory in place of rotating media but otherwise appears as a magnetic disk. It fits in the hierarchy between main memory and DASD. The solid state disk is an example of a product which fills a specific need and fits well into the hierarchy.
The next level in the hierarchy is occupied by rotating disk memory, that is, Direct Access Disk Storage (DASD). Most applications programs use DASD for I/O because it is fast and readily accessed. DASD and all lower levels of the hierarchy are accessed via the IBM I/O Channel.
The evolution of disk storage for IBM mainframes follows key technology improvements. IBM Model 2311 was announced in 1960 with conventional head technology. The 3330 drive introduced in 1970 for the IBM 360 computer moved the heads into the disk pack with improved reliability. IBM Model 3350 was announced in 1973 with Winchester head technology. The 3350 represents the first true sealed Winchester type drive. The clean environment allowed lower flying heights and higher density and speed. IBM Model 3370 was announced in 1973 with Whitney head technology. The 3370 was the first thin film disk. Made in 1979, thin film improves performance by improving head and media performance, not by lower flying heights. The bit density of the 3370 and 3380 is higher than the older Winchester drives.
DASD does have disadvantages. Because of its rewritable nature, data can be lost. The tight tolerances which make the media dense and fast also make it susceptible to head crashes and lost data. Magnetic disks are required to be backed up onto magnetic tape in order to restore data after a failure. This is one of the primary uses for magnetic tape.
DASD uses non-removable media. In order to maintain the tolerances needed for high density it was necessary to build a unified head-disk-assembly integral to the subsystem. Data must be transferred to another media if it is to be stored offline, i.e. aged data must be transferred to magnetic tape. This is another primary use for magnetic tape. DASD is expensive. Although the cost per megabyte of online storage has decreased steadily over the years, users must still remove aged datasets.
The close tolerances required to achieve high density and high speed make the disks susceptible to environmental fluctuations. DASD is physically large, requiring 2 or 3 tiles for a five giga byte system, plus the required free space for maintenance access. The 3370 and 3380 use different recording formats but can share the same type 3880 storage controller. These two formats are important to the introduction of optical storage for IBM mainframes. The two competing formats are Fixed Block Architecture (FBA), used in the 3370, and Count Key Data, used in the 3380.
Like Count Key Data DASD, the FBA disk drives provide high speed random access to datasets. In general, the performance of the 3370 devices is equivalent to their Count Key Data counterparts. The distinction of FBA is the way in which datasets are recorded on the disk and their locations managed. For fixed block architecture devices, tracks and records are formatted at the time of manufacture. The user does not specify the length of the data area. Records are specified for selection by relative block numbers. The device is formatted into a continuous sequence of numbered blocks and arranged at evenly spaced sector locations. The subsystem converts the block number to the track address and sector location. The 3380 is a Count Key Data drive. With count, key and data device types, the programmer decides the length in bytes of each physical record. The I/O Channel programmer writes certain control information for each record in the pattern that is predefined for the count, key, and data format. All count, key, and data tracks are formatted beginning at the index point and ending at the following index. Each track has the same basic format: track home address and track descriptor record, followed by user records. One or more user records, with count, key and data, are written following the descriptor record. Each of the Count, Key and Data areas is separated by a gap. The count area contains the location of the data area that follows. The location is specified by the track address (cylinder and head numbers) and record number. The count area also specifies the length of the key and data areas. The optional key area is used to identify the data, while the data area contains the user's logical records. The number of records on a track vary according to the length of each of the data areas. With count key data devices the channel sends a seek command to position the access mechanism to a track. Once on track the desired record must be located. This is accomplished by a command that causes a search for the record identifier. The desired record identifier is compared to the record numbers in the count area of records on the track until a match is achieved.
Even though FBA does not allow users the freedom to decide where on a disk surface information will be stored, some argue that FBA is more efficient than count key data because it reduces the amount of disk space required for housekeeping functions. While count key data allows users to place their most active files where they can be quickly accessed, the housekeeping for count key data can be unwieldy, and the small gaps that count key data requires between fields consume storage capacity. Eventually the gaps can actually occupy more space than the data. Like their FBA counterparts, the 3380s have the advantages of high speed and update in place. They also share the deficiencies of the FBA drives by the requirement for backup to nonvolatile media, the lack of removability of the disk pack itself, high cost, stringent environment and large physical size.
The next level in the hierarchy is occupied by magnetic tape. Tape has seen fewer changes over the years than disk. Chief among the advantages of tape are its standard format and transportability. The media is removable and can be shipped from place to place. The media is also inexpensive.
Tape does have a number of significant disadvantages. Access is limited to serial searching of the media. For this reason most data is transferred from tape to disk prior to processing. The data also cannot be easily updated in place, a record once written cannot be replaced by a record of different length.
Another disadvantage is that the tape media is expensive to store because it requires a controlled environment to halt deterioration. Deterioration of magnetic media is a major concern in archiving. Several mechanisms contribute to the deterioration of media over time. Chief among these are the aging due to humidity and stress temperature changes. Humidity has the effect of weakening the tape and allowing the magnetized portion to separate from the backing. Changes in temperature cause the tape and the hub to change size at different rates. This causes stress in the pack. It is often advisable to periodically rewind tapes to relieve stress.
Perhaps the most limiting disadvantage of the tape is the amount of time lost in manual operations. Manual operations on magnetic tape are a major source of errors in computer centers. In general, the operators are required to locate, mount and store tape reels based on the six digit volume-serial number (VSN) located on the case. Transposition of numbers, mislabeling and misfiling often lead to mounting of the wrong reel. In a labeled environment, the tape management program will catch such errors and prevent further processing. An amend will occur and the job will be rerun at a later time. In a no label or bypass label environment, data loss will almost always occur with highly unpredictable results, sometimes not evident at the moment.
The 3420 reel tape line has been the standard for years. Because of the stability of the technology, it has been relatively easy for the plug compatible manufactures to copy the 3420. The longevity of the nine track tape is in part due to the industry standardization on this format. Data written on one machine can almost always be recovered by another machine even when different manufacturers, e.g. when Digital Equipment Corporation (DEC) and IBM, tapes are involved. There are of course issues of conversion of code (ASCII vs EBCDIC) and number representation, but in general the data can be read into the machine to make such conversions possible. The long history of these drives makes them an obvious choice for many tasks requiring long term storage.
The performance of the 3420 reel tape drive has increased over the years. The 3420-8 is a two-hundred inches per second machine which at six-thousand-two-hundred-and-fifty bits per inch, transfers data at one-point-two-five mega bytes per second. However, the transfer speed is deceptive since it is influenced by and sometimes swamped by other factors. Chief among these are the mounting and dismounting manual operations and sequential access inherent in a tape mechanism. A typical mount or dismount requires time. A rewind may occupy sixty seconds. Access to the tape itself, if it is in an archive for instance, may be measured in the tens of minutes. The lost time is lost channel time and lost drive time. Drives can be viewed as relatively inexpensive devices whose time can be wasted. Channel time is more precious because the channel is a shared resource and slow performance can impact other tasks. Thus, the most important loss of tape performance comes from sequential access and interblock gaps.
An interblock gap (IBG) separates two datasets on a tape. These IBGs are required to allow an area for the drive to come to rest between reads and writes. The IBGs consume both time and storage. There is a relationship between the effective transfer data rate expressed in mega bytes per second and the block size expressed in kilo bytes. The transfer rate rapidly increases to a maximum rate as the block size increases. The IBG for 3420 drives is the equivalent to a two kilo byte block. The actual transfer speed of the drive drops to one-half when the block size is reduced to the IBG length.
The 3480 cartridge tape drive advantages are firstly the cartridge itself, which is small in size and totally protected, and secondly the high speed and small size of the transport. The 3480 offers several advantages over of the reel system from the operator's point of view. The cartridge is smaller, thus occupying less shelf and cart space. The tape is fully protected in the cartridge and is never exposed or handled. The machine is fully autoloading and is available from IBM with a simple loader. The drives include a large display which is used to cue the operator for tape mounts and dismounts and to indicate drive status.
The 3480 achieves sustained three mega bytes per second transfer speeds through its high density eighteen track recording system. This is an important feature in backup situations since it allows speed matching between DASD and tape. As the number of strings of DASD increases, the amount of data required to be backed up on tape increases as well. This is in general a daily process, and ties up both I/O Channel time and operator time. The 3480 cartridge is designed to ensure that it will hold the entire contents of a twenty-four-hundred foot reel to make conversion easier.
Like the 3420 reel, the 3480 cartridge does need to be rewound and mounted. These operations are time consuming. If speed is the issue, then rewind time must be considered since it occupies the time of the drive (although not the channel). At three mega bytes per second, a two-hundred mega byte tape requires sixty-seven seconds to read. Nearly this same amount of time will be required to rewind this tape. The total drive data rate drops by one-half. Load and unload time amounts to about five seconds each. Compared to the rewind time this is small. If only a small portion of the tape needs to be read, then the load, search, rewind, unload times can be significant. Like the 3420, the 3480 uses an interblock gap to separate data records. The IBG length for the 3480 is four kilo bytes, twice that of the 3420.
The 3480's advantages over the reel 3420 are many. It is more transportable than the equivalent reel capacity and is better protected. The tape format includes a new block id function which will allow rapid access to data areas with little CPU involvement. The operator functions are much improved, with a display system which prompts the operator to mount and dismount cartridges. A path selection function is provided to assist in management of shared tape. Finally, the cache speeds access to data by keeping many tape motion delays isolated from the CPU and I/O Channel.
The disadvantages of tape media do carry over into the 3480 cartridge. It remains a sequential access mechanism with the difficulty of updating in place, and storage costs remain high due to the environmental restrictions on the media itself. The 3480 specification are generally: A one-half inch, eighteen track, high density tape that is enclosed in a compact cartridge for protection and automatic threading; Small tape cartridges four-by-five-by-one inches with storage capacity of two-hundred mega bytes; A tape drive that moves the tape without the need for capstans or vacuum columns; A fast search capability that allows a program to position a tape to a specific block without constant processor or control unit supervision; A message display on each tape drive to present operating system messages to the operator; Reduced physical and mechanical complexity; Superior error correction and better reliability; and Data transfer rates of up to three mega bytes per second.
The two dominant hardcopy mechanisms occupy the bottom of the hierarchy. Paper is the only true archival media. This stems not from the character of the media itself but from the character of the machine required to read the media. It is assumed that any media written more than twenty years prior will be unreadable because no machines will be available to read it. The use of paper will remain high for this reason. It is readily used, inexpensive and uniquely convenient in that it can be altered by the user.
Microfilm enjoys the advantage of paper in that with simple optics it can be human readable. Microfilm can be considered to be true archival media. Unfortunately this characteristic is also a disadvantage because, in general, what is human readable is not machine readable. Microfilm is consequently an output media, not readily transferred back to the machine for further processing. It is not as convenient as paper because it requires a machine for magnification and cannot be altered by the user. However, it is readily stored, and conveniently indexed for retrieval.
Several alternate storage systems have been made available. Each system is designed to occupy a niche in the hierarchy. StorageTek announced their 4400 Automated Cartridge System, a fully automated, 3480 cartridge based information storage and retrieval system. StorageTek expects the 4400 to fill the gap between online DASD and offline tape, and has coined and trademarked the term Nearline to describe their product. The StorageTek 4400 is an important offering because it uses the 3480 cartridge. The IBM 3850 Mass Store System suffered from the fact that it used a unique media requiring conversion from tape or access from DASD. StorageTek has eliminated this objection by using the 3480 cartridge. The 4400 uses a bar code system to label cartridges. These bar codes are read by a robotic picker. StorageTek offers a software package to provide for a full access mechanism.
The IBM 3850 Mass Store System provided up to four-hundred-and-seventy-two giga bytes of storage. The data was stored on unique cartridges, and when requested by the processor was transferred to DASD for processing. The processed data was then transferred back to the cartridge for storage. Cartridges were stored in a honeycomb arrangement and accessed via a picker. First the host computer issues commands to a Staging adapter for accessing data on DASD which conform to the command set for the IBM 3330 Disk storage devices. The staging adapter in connected to both a DASD and a cartridge storage. The staging adapter transfer information on the cartridge storage to the DASD storage for host computer use. The host computer sends Mass Storage System instructions to a Mass Storage Control Unit for data cartridge selection. The DASD storage stores Mass Storage Control tables for look up to the data cartridge. The Mass Storage System provides cartridge tape storage but appearing to the user as conventional DASD storage.
Tape, contained in the data cartridges, was used for data storage, while disk was used to make the data available to the CPU for processing. The storage, retrieval and management of the data in the cartridges was handled by the mass storage system and was transparent to the user. User programs operated as if the data always existed on disk. Introduced over eleven years ago the system was not a success. Early problems with microcode and users skepticism about the cartridge combined to kill the product. No further work on the 3850 is planned. The major contribution of the 3850 was the software developed to support it. The Hierarchical Storage Manager (HSM) developed for the 3850 forms the basis for much of the storage management software available now.
Remote Storage has been used at user terminals. A trend in storage which should be considered is the distribution of storage to user networked personal computers and low end mainframes. Significant software will be required to allow efficient well managed distributed storage. The IBM storage management of datasets as they migrate between disk, tape, and mass storage devices is becoming a major problem. Users will be able to do it only with help from the system itself. IBM and independent software competitors are hard at work preparing products that transparently and automatically arrange data files in hierarchical schemes, putting the most active ones on disk for instant access and keeping less-used files close at hand on cheaper storage devices.
IBM has announced the Data Facility family of program products. IBM is evolving to a system-managed storage environment based on Data Facility. These products include the Data Facility Product, the Data Facility Hierarchical Storage Manager, Data Facility Sorting package, and the Data Facility Data Set Services program. The Data facility family works with other IBM products for data security, timesharing, and relational database management. IBM announced a common interactive interface for the Data Facility family in the form of the Interactive Storage Management Facility. Based on the facilities of IBM's Interactive System Productivity Facility, the Interactive Storage Management Facility provides storage administrators and users with a series of interactive screens to choose from. They can also select datasets for certain operations. One of Interactive Storage Manger Facility's prime advantages is that it eliminates the need for creating and debugging Job Control Language code. Interactive Storage Manager Facility users can also review, copy and delete datasets. The Interactive Storage Manager Facility product provides a simplified syntax, prompts, defaults, help, and other features to improve productivity and to shield users from much of the complexity of the individual Data Facility programs.
IBM's Hierarchical Storage Manager keeps track of datasets as they move among disk, tape and, initially the 3850 mass storage device. Hierarchical Storage Manger staged datasets on disk after they were called out of the 3850 by an application program. Archive Storage Manager was enhanced with new facilities that enable it to better track the usage of individual datasets, storing them on the appropriate mechanical devices. With the Archive Storage Manager, users can define classes of datasets that are to be treated in different ways. The goal is to free disk space for high-priority data and to improve overall system and worker productivity. The user has no need to know where his data is because the Archive Storage Manager keeps track of where each dataset is and can move it as needed.
As peripheral controller software becomes more intelligent, and new storage and memory technologies add more levels to the overall storage hierarchy, IBM would like to manage storage at the logical level with all datasets appearing to reside on a single device.
Operating Systems need to have tape access. Understanding how the mainframe gets access to a particular data set is important when considering an alternate storage mechanism. Considering tape in particular, the task is to resolve the location of a particular dataset. In general this process involves consulting a catalog, maintained by MVS or an applications program, and determining the volume-serial number (VSN) of the tape which contains the dataset. A mount request is then generated to prompt an operator to mount the desired VSN. The system will then read the label on the tape to verify that the correct VSN has been mounted. This accomplished the job requiring the dataset located on the VSN will be run. The process has many variations depending on the operating system, the applications programs and the operating procedures of the individual data center. Under MVS with a package like UCC-One which is a product of UCCEL Corporation that solves tape management problems, the control is quite strict and should be error free. In a Vault Management environment the control is less automated and may be done primarily by manual cataloging. Each data center operates according to its own policies and procedures in this area.
There is a market place for optical storage devices. Within the storage hierarchy and the devices which implement it, there exists a place for WORM optical storage. Just as the solid state disk memory fits certain unique applications, WORM optical media fits certain applications. Optical storage is not expected to replace any existing storage mechanism. It will augment existing mechanisms and improve total system performance. So too, erasable optical has its place and will not displace WORM. Each storage media has its positive and negative aspects, as does WORM optical storage.
There are many reasons to use WORM optical storage in certain applications. On the positive side, optical offers high storage density, permanence, removability, long life, and low media cost. The high density of optical is achieved by using laser energy to read and write the media. The important observation is that this density is not achieved with the very low mechanical clearances of magnetic media. Head crashes are not a concern with optical, and neither is head/tape wear. Because the media is removable and easily stored, it is ideal for archiving data in place of tape, especially data that is of permanent long term value. Once written, the data is permanent and cannot be erased or rewritten. This is an important consideration for certain record keeping tasks. The optical storage media is inherently tamper-proof.
Optical storage media costs today are roughly twice that of tape media on a per megabyte basis. This assumes that tape is one-hundred percent utilized which is rarely the case in IBM systems. Media costs are expected to continue to decrease. As with other media, what is an advantage on one side is seen as a disadvantage on the other. This is true of optical as well. There are several criteria which mitigate against the use of optical media in the IBM environment. Perhaps the-most significant to date has been the lack of a complete turnkey solution. Some frustrated users have resorted to their own implementations using Direct Access Control Units, and similar methods. This aspect of the problem is solved by the introduction of the present invention.
The write speed is the limiting factor because of the amount of energy required from the laser to mark the media. Higher revolutions per minute require higher laser power and reduce the mean time between failures. Write speeds are also influenced by the error correction mechanism. If a separate revolution is required to verify the data then the write speed can suffer due to rotational latency. If the correction is done on the fly then the error correction circuits must decide quickly if an error occurred sometimes resulting in unnecessarily rewriting sectors. Read speeds could be improved by creating a read only transport which spins at higher speed.
Applications which are suitable for optical storage include those which can be satisfied with a low speed device, and involve the storage of long lived data, and involve the storage of large volumes of data. Where all three factors are present optical is a good fit. Applications which are unsuitable for optical include those which require a high speed device, OR involve storage of short lived data, or involve storage of small amounts of data. Where any one of these three factors is present optical is a poor fit.
Many applications often cited for optical simply are out of reach with today's technology. The oft cited seismic data application is ideal from many respects: it is very high quantity, of permanent archival value, and write once. Unfortunately the collection of this data occurs at speeds in excess of one mega byte per second. The seismic data application is not a suitable application because of the high speed required during data collection. DASD backup fails the condition because the data is short lived. Backup tapes are generally retained for a short period, beyond which they lose their value.
Advantages of WORM optical are high density, permanent, tamper-proof, stable media and removable media. Disadvantages of WORM Optical are low speed, high media cost, write once and a lack of standardization.
A new optical controller could take full advantage of the optical storage mechanism, in particular the ability for random access to very large data bases. To accomplish these objectives the controller designer would define a set of I/O CCWs (Channel Command Words) commands for this new device. Seek commands would be included to achieve random access. A directory structure, like a Volume Table of Contents, would be defined which would conserve media. Ultimately, even a new access method would be defined and new host software written to utilize the new device. However, this software development would be a major undertaking. Several man years would be invested in developing software to use the new device. This new software represents a significant risk for vendor and customer alike. It is this software issue which motivates an emulation to interface the optical disk. An emulation has the advantage of existing software support from IBM, with the disadvantage that the emulation cannot make the best use of the new device's unique characteristics.
DASD emulation is the most desirable emulation because most applications software is written to utilize DASD. DASD is organized to allow efficient random access to large datasets. This is accomplished through the use of individual volumes set aside for each dataset. Consider the Volume Table of Contents used in count key data devices. Each volume contains a Volume Table of Contents of the datasets and of the unused space on the volume. The Volume Table of Contents is used to control the allocation of space on the volume, and to determine where a dataset is located. Within the Volume Table of Contents are records which name and describe individual datasets. The descriptions include the cylinder and head location and number of extents. When a dataset is written, the records in the Volume Table of Contents describing the dataset are updated to activate the new dataset record, and to reduce the free space remaining on the volume. It is this updating process which is most difficult to handle efficiently with WORM media. One possible solution to the disk emulation is to stage data on magnetic or solid state memory prior to writing to optical disk. In this approach a volume would be held in staging memory until it was closed and then written to disk. The staging memory could be in main memory, as done by Perceptics' attachments on DEC computers. This requires special software and consumes main memory. This staging memory could be of considerable size, essentially a solid state disk acting as the front end cache to the optical disks. This is an expensive solution.
To solve the problems of DASD emulation a complex system is required. The problems include backup for the volatile staging memory with error recovery mechanisms. Emulating count key data drives would be excessively consumptive of media. Emulating FBA drives would be a better fit since the optical drives with SCSI (Small Computer System Interface) interfaces are FBA oriented. However, the future of FBA is uncertain. The added flexibility afforded to the user by DASD emulation is probably not worth the risks and cost. The eventual availability of erasable optical may simplify this problem, DASD may then be the emulation of choice.
Emulating the IBM 3850 Mass Storage System could be a reasonable choice because software exists to handle this machine. However, because only a few were sold, this software may not be supported. A vendor could find that the essential software was no longer available or had been modified. This seems an untenable situation. Emulating the StorageTek Automated Cartridge System ACS4400 also appears risky. Again the availability of software and the assurance of continued compatibility with IBM cannot be assured.
Most turnkey optical systems offered are configured as document servers with an interface such as Ethernet or RS232. Optical is used to store text and images, either coded or scanned, and access is primarily by end users at terminals who wish to view, copy, or edit the material. A similar approach could be used in an IBM environment by emulating a controller, the 3274 or 3174. For document handling this is a viable approach. The Software required to utilize such an emulation would be extensive.
Emulating a tape drive is another choice. Unfortunately tape is most often used for DASD backup and for storage of aged data. In the IBM world only limited applications expect to work with data directly on tape. Typically data is first transferred to DASD, manipulated, and later returned to tape. From one aspect this is good because it makes the WORM nature of optical less objectionable. From the second viewpoint this is bad because it means that the random access nature of the optical media is rarely if ever used; data is streamed sequentially to DASD and accessed randomly once loaded. The process of moving data between tape and DASD consumes I/O Channel resources. Unfortunately this highlights a current weakness of optical with a relatively slow read/write transfer rate.
However, features available on the new 3480 drives make tape an attractive choice since they offset some of the weaknesses noted above. In particular, the 3480 associates a logical block id with each record. This logical block id permits searching for specific records without the need for a series of forward space file reads. Few programs that take advantage of the logical block ids use the tape drive offline searching capabilities. This may change as 3480s become more widespread. The 3480 also includes a cache memory. A similar concept can be used in optical to speed transfers across the I/O Channel.
The current IBM System 370 storage hierarchy consists of main memory, disk memory or Direct Access Storage Devices (DASD), reel and cartridge magnetic tapes, and hardcopy such as paper and microfilm. Within the retrieval levels of this hierarchy, only data stored on DASD is available at all times for online mainframe processing. Data recorded on magnetic tapes is generally stored offline. Major time delays are frequently associated with the transfer of tape data from the tape library onto DASD. DASD storage is fast but remains expensive, while tape library storage suffers from both slow access and the attendant media storage space constraints and costs.
Consequently, a need has existed for a compact form of high capacity storage that is retrievable online when additional use of the data in mainframe processing is required. Optical storage can satisfy this compact, high-capacity online storage need. By emulating 3480 cartridge tape subsystems, Write Once Read Many (WORM), non-erasable, removable, optical disk cartridge storage can be made conveniently available to IBM and compatible mainframes. The present applicant choose to emulate the 3480 tape subsystem. This is the most general purpose device, able to satisfy the broadest set of requirements. The present invention solves or reduces the disadvantages of the prior art mass memory systems.